Published Paper Details:

Abstract-- In order to enhance signal processing, this work explores the integration of neural networks (NN) at very-large-scale integration (VLSI) levels. In terms of SNR, processing time (latency), energy consumption, and accuracy of signal processing tasks, we suggested and implemented a neural-network-based VLSI based on NNs and compared it with a traditional DSP. Our findings show that, in comparison to DSP techniques, the NN-based circuit efficiently boosts SNR by up to 7 dB. According to latency measurements, the NN-based circuit outperforms the DSP, which has average latencies of 3.5 and 6.2 microseconds for single and complex jobs, respectively, with average latencies of 1.2 and 2.8 microseconds. Additionally, with a 45 mW power usage, the NN implementation reduces the average power by 40%. In contrast, DSP uses 75 mW. With 98.2% accuracies for signal filtering, 97.5% for denoising, and 99.1% for pattern recognition, NN-based circuits outperform DSP, which has accuracies of 95.7%, 92.3%, and 96.8%, respectively, according to accuracy study. These results highlight the better precision, throughput, and energy efficiency that the NN-based circuit can provide. In conclusion, neural network implementation on VLSI can offer notable benefits over DSP-based implementation, making it seem like a promising technology for advanced signal processing that demands high precision, high efficiency, and high performance. By using NN-based circuits in general signal processing applications like communications, imaging, and real-time data analysis, this study lays the groundwork for future advancements in VLSI.